A method for path planning of a six-legged crawler composite robot

Abstract

The application discloses a kind of six-foot track composite robot path planning methods, comprising the following steps: S1: establish environment map, construct binary map and three layers of passability semantic grid;S2: initial global path is generated using improved RRT algorithm, including the introduction of improved artificial potential field guide and adaptive step adjustment;S3: initial global path is mapped to three layers of passability semantic grid, and path semantic is labeled and segmented, and track section and six-foot section are obtained;S4: whether there is track bypass channel, yes, then evaluate time cost;S5: calculate six-foot leap and track bypass time cost, and according to the decision rule of band hysteresis margin, decide to use leap or bypass;S6: decision is bypass, then generate bypass path section and splice with global path;S7: output final planning path.The application can effectively reduce global path length, planning time consumption, sampling frequency and turning point quantity, generate more compact, more smooth initial global path.

Description

Technical Field

[0001] This invention belongs to the field of mobile robot path planning technology, specifically a path planning method for a six-legged tracked composite robot. Background Technology

[0002] In unstructured environments such as outdoor factory areas and disaster response sites, terrain features such as flat roads, low obstacles, ditches, and broken debris are often distributed alternately, placing higher demands on the robot's mobility and passability. Compared to single-mobility platforms, hexa-legged tracked hybrid robots combine the continuous propulsion capability of tracked mode with the ability to traverse complex terrain in hexa-legged mode, thus possessing strong adaptability potential in such scenarios. However, the path planning problem for this type of platform is not equivalent to the geometric obstacle avoidance problem of traditional mobile robots. Since different motion modes exhibit significant differences in speed, passability, and stability, and mode switching requires additional time, the planning results must not only ensure global path reachability but also reflect the passability attributes of different terrain sections and their corresponding mode adaptation relationships.

[0003] Path planning aims to generate a collision-free feasible path from a starting state to a target state within the workspace, based on the robot's perception of the environment and its own motion constraints. It is a crucial component of autonomous robot systems. Existing global path planning methods can be broadly categorized into graph search-based methods and sampling-based methods. The former typically involves first discretizing or modeling the environment as a graph, then searching for the optimal or suboptimal path from the starting point to the destination within the graph. Typical examples include Dijkstra's algorithm. and Algorithms such as [list of algorithms]. These methods have a clear search process and relatively deterministic results, making them well-suited for regular grid maps or structured environments. In recent years, related research has continued to advance. For example, some studies have proposed fast path planning methods for large-scale safety box sets, combining weighted graph shortest path search with convex optimization to achieve rapid generation of safe and smooth trajectories. Simultaneously, researchers have begun to explicitly introduce information such as risk and drivability into graph structures to enhance the ability of path search to express complex environmental attributes. However, graph search-based methods typically rely heavily on the discrete quality of the environment and the results of graph structure construction. As the dimensionality of the search space increases or the continuity and complexity of the environment increase, their computational burden also increases. In contrast, planning algorithms based on random sampling have received widespread attention due to their strong adaptability to high-dimensional spaces and complex obstacle environments. The (Rapidly-exploring Random Tree) algorithm is highly representative. By randomly sampling in the state space and incrementally constructing a tree structure to achieve connection between the origin and destination, when feasible paths exist, as the number of samplings approaches infinity, the probability of finding a solution by this algorithm approaches 1, thus exhibiting probabilistic completeness. Meanwhile, The algorithm has a simple structure and low dependence on explicit environmental modeling, thus it has a strong application foundation in global path planning in complex environments.

[0004] However, the foundation The algorithm still has significant limitations. Because its expansion process heavily relies on random sampling, it easily generates a large number of invalid samples, leading to a decrease in search efficiency. Furthermore, the generated paths often suffer from numerous redundant segments, frequent turns, and insufficient smoothness, especially in scenarios with dense obstacles or narrow passages. To improve the efficiency of solving for the initial feasible path, researchers proposed... The algorithm accelerates the search process through the mutual growth of two trees and a greedy connection mechanism, but this method mainly focuses on obtaining feasible solutions and does not adequately consider path cost optimization. Subsequently, The algorithm continuously optimizes path cost through node reselection and local reconnection mechanisms. The algorithm improves the effective sampling rate by constraining the sampling space. Besides sampling space optimization, some studies have also introduced methods such as artificial potential fields to enhance the guidance of the expansion direction. While artificial potential field methods can utilize local gradient information to improve the search direction, they are prone to getting trapped in local minima. To balance global exploration and local utilization, some studies have introduced APF (Advanced Persistent Field) methods. The framework proposes Algorithm, in addition , and These methods have also been improved in terms of node reconnection, sampling selection, and convergence process optimization. Algorithms. Overall, existing... Its variants mainly focus on optimizing sampling efficiency, search speed, path length and convergence performance, providing an important reference for the initial global path generation in complex environments. However, they do not adequately consider the differences in the passage of different motion modes in hexa-legged tracked composite robots, thus making it difficult to directly support mode selection and path correction in local segments.

[0005] On the other hand, recent research on mobile robots and composite platforms in complex terrain has shown that robot path planning is no longer simply a geometric obstacle avoidance problem, but is closely related to platform configuration, terrain attributes, and mobility strategies. One type of research focuses on composite platforms such as wheel-legged, leg-wheel-tracked, and articulated tracks, improving the robot's mobility and environmental adaptability in unstructured environments by integrating multiple motion configurations. Another type of research addresses complex terrain navigation, introducing methods such as terrain perception, risk perception, and semantic segmentation to incorporate terrain drivability, environmental uncertainty, and safety risks into the path planning process, thereby improving the robot's planning robustness and mobility rationality in complex terrain environments. These studies indicate that the planning results for robots in complex terrain need to further reflect terrain drivability, platform adaptability, and strategic decision-making information, and cannot remain solely at the geometric connectivity level. However, the aforementioned research focuses more on wheel-legged robots, articulated tracked robots, or general complex terrain mobile robots, with relatively insufficient attention paid to the path planning problem of hexa-tracked composite robots. Especially after the initial global path is generated, it is still necessary to further combine environmental semantic information and platform mode differences to label the passage attributes and structure each section of the path, and to judge the time cost of crossing and detouring strategies for local sections such as low obstacles and ditches, so that the final path result is more in line with the actual passage requirements of the hexa-legged tracked composite robot.

[0006] In summary, the existing technology has at least two shortcomings: First, at the initial global path generation level, the basic... Furthermore, some improved methods still suffer from problems such as excessive invalid sampling, low search efficiency, and insufficient path quality, making it difficult to stably generate compact, smooth, and suitable initial global paths for hybrid robots. Secondly, in terms of path processing for hexa-legged tracked hybrid robots, existing methods do not adequately consider the differences between tracked and hexa-legged modes in terms of terrain adaptability, motion efficiency, and mode switching costs, resulting in the initial global path being unable to directly support mode selection and correction in local segments. Summary of the Invention

[0007] To address the problems existing in the prior art, the present invention aims to provide a path planning method for a hexa-tracked composite robot, which can effectively reduce the global path length, planning time, number of samplings and number of turning points in both random and factory environments, generating a more compact and smoother initial global path.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A path planning method for a hexa-legged tracked composite robot includes the following steps: Step S1: Establish an environmental map, construct a binary map and a three-layer drivability semantic mesh; Step S2: Employ an improved... The algorithm generates an initial global path, and the improvements include introducing an improved artificial potential field for guidance and adaptive step size adjustment; Step S3: Map the initial global path to the three-layer drivability semantic grid, perform semantic annotation and segmentation on the path to obtain tracked segments and hexapod segments; Step S4: For the hexapod segments, determine whether there is a tracked detour channel. If so, proceed to time cost evaluation; otherwise, maintain the original path; Step S5: Calculate the time cost of hexapod crossing and tracked detour respectively, and decide whether to cross or detour based on the discrimination rule with hysteresis margin; Step S6: If the decision is to detour, generate a detour path segment and concatenate it with the global path to replace the original hexapod segment; Step S7: Output the final planned path.

[0010] As a further improvement to the above technical solution:

[0011] The improved artificial potential field guidance described in step S2 is as follows: the attraction of the target point, the attraction of the random sampling point, and the repulsive force of the obstacle are all included in the calculation of the expansion direction, and the resultant force direction is used as the new node generation direction.

[0012] When there is a problem of unreachable target, a distance factor between the robot and the target point is introduced into the repulsion function.

[0013] The adaptive step size adjustment in step S2 is as follows: dynamically adjust the extended step size according to the distance between the current node and the nearest obstacle, so that the step size is reduced in narrow areas and increased in open areas.

[0014] The dynamic adjustment adopts an adjustment model based on the Sigmoid function, with a step size range of 0.5 to 2 times the fixed step size.

[0015] The semantic annotation and segmentation in step S3 includes: assigning track, leg, or blocked labels to path points, dividing the area into segments according to the continuity of the labels, the segment where the track label is located is the tracked passable area, the segment where the leg label is located is the six-legged passable area, the segment where the blocked label is located is the area that is impassable in both modes, and setting a mode switching pre-switching interval at the boundary of the six-legged passable area.

[0016] The time cost of the six-legged crossing scheme described in step S5 Time cost of track bypass scheme for:

[0017]

[0018]

[0019] in, This refers to the time consumed during a single mode switch. The equivalent length of the global path for areas where low obstacles or ravines can be crossed. For six-legged mode speed, ∈{1, 2} represents the number of switching operations; if the tracks are restored immediately after crossing, then... =2, otherwise 1. For local The algorithm generates the length of the detour segment. This refers to the track speed.

[0020] The hysteresis margin discrimination rule mentioned in step S5 is as follows: Let As a hysteresis margin, ,like If the vehicle chooses to go around and maintain tracked mode, it will choose to cross on six legs.

[0021] The generation of detour path segments in step S6 adopts a local approach. The algorithm plans within the passable space of the track. The splicing includes advancing along the original global path to determine candidate sub-target points, and determining the final reconnection position after combining semantic attributes, collision status and reachability filtering, so that the local detour path segment can avoid the original local segment and rejoin the original global path.

[0022] If a feasible path to the sub-target point cannot be generated when generating the detour path segment, the detour is determined to be infeasible, and the crossing path corresponding to the original local segment is retained.

[0023] The beneficial effects of this invention are:

[0024] (1) This invention introduces an improved artificial potential field guidance mechanism, which incorporates the attraction of the target point, the attraction of random sampling points, and the repulsive force of obstacles into the calculation of the expansion direction, and combines it with adaptive step size adjustment (dynamically adjusting the step size based on the Sigmoid function), effectively overcoming the limitations of the basic... The algorithm suffers from drawbacks such as excessive invalid sampling, strong blind search, and redundant and uneven paths. Simulation experiments show that, in both random and factory environments, this invention improves the global path search efficiency and path quality. The method described in this invention (…) The global path length, planning time, number of samples, and number of inflection points of the sample are all significantly lower than those of the basic sample. Compared with other algorithms, the algorithm generates a shorter, smoother initial global path with fewer turns, which is beneficial for subsequent path segmentation and mode switching.

[0025] (2) This invention constructs a three-layer drivability semantic mesh, dividing the environment into tracked drivable areas, hexapod drivable areas, and areas where neither is drivable. Semantic information is mapped to the global path, realizing path semantic annotation and structured segmentation, and achieving effective characterization of the drivability characteristics of the composite robot. This processing ensures that the path results not only include geometric connectivity but also directly reflect the drivability attributes and motion mode adaptation relationships of different terrain sections, providing a unified environmental and path description basis for the mode switching and local decision-making of the hexapod tracked composite robot.

[0026] (3) This invention establishes a time cost evaluation model for crossing and detouring in response to the differences in motion modes of the hexa-legged tracked composite robot. It comprehensively considers factors such as mode switching time, low-speed passage in hexa-legged mode, and detouring distance in tracked mode, and introduces a hysteresis margin discrimination rule to avoid frequent mode switching in critical states. Simulation results confirm that this mechanism can make reasonable decisions for different local segments—using tracked detouring and partially replacing the original path when detouring time cost is lower, and retaining the crossing path when crossing is better. It reasonably solves the discrimination problem of crossing and detouring strategies in local segments, so that the final path result is more in line with the actual passage efficiency of the composite robot.

[0027] (4) Compared with the pure tracked mode, this invention fully utilizes the traversing capability of the hexa-tracked composite mode, expanding the effective passable space and significantly reducing the global path length, sampling times, and planning time. Simultaneously, the local correction mechanism... Alternative path segments are generated and concatenated with the global path, replacing only local segments without altering the overall direction. A rollback mechanism is implemented, thus achieving local strategy optimization while maintaining global structural stability. Experiments show that the method of this invention can stably generate compact, smooth, and pattern-adaptive paths in both random and factory environments, significantly improving the task completion efficiency of hexa-legged tracked composite robots and their planning robustness in complex unstructured environments. Attached Figure Description

[0028] Figure 1 A flowchart for the fusion path planning of a hexa-legged tracked composite robot.

[0029] Figure 2 This is a binary occupancy grid diagram for the present invention.

[0030] Figure 3 This invention provides a three-layer accessibility semantic grid.

[0031] Figure 4 For the present invention Algorithm extension graph.

[0032] Figure 5 For the present invention Algorithm flowchart.

[0033] Figure 6 For the present invention A force diagram.

[0034] Figure 7 When the objective of this invention is unattainable A force diagram.

[0035] Figure 8 This is a comparative schematic diagram of the adaptive step size of the present invention, wherein (a) is a schematic diagram of fixed step size expansion and (b) is a schematic diagram of adaptive step size expansion.

[0036] Figure 9 This is a schematic diagram of path mapping and segmentation according to an embodiment of the present invention.

[0037] Figure 10 This is a solid image of a hexa-legged tracked composite robot.

[0038] Figure 11 This is a 300 m × 200 m random environment map according to one embodiment of the present invention.

[0039] Figure 12 This is a 400 m × 200 m factory area environmental map according to an embodiment of the present invention.

[0040] Figure 13 This is a schematic diagram of path correction in a random environment map according to an embodiment of the present invention.

[0041] Figure 14 for Figure 13 Enlarged schematic diagrams of the local correction region and correction candidate segments, where (a), (b), (c), (d), and (e) are respectively... Figure 13 Enlarged schematic diagrams of points a, b, c, d, and e.

[0042] Figure 15 This is a comparison chart of different algorithm paths for the same composite mode in a random environment map.

[0043] Figure 16 This is a comparison chart of different algorithm paths using the same composite mode for the factory environment map.

[0044] Figure 17 This is a comparison chart of the overall performance of a random environment map under different motion modes.

[0045] Figure 18 This is a comparison map of the overall performance under different motion modes in the factory environment map. Detailed Implementation

[0046] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0047] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "on top of" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatial relative descriptions used herein will be interpreted accordingly.

[0048] A path planning method for a hexa-legged tracked hybrid robot, such as Figure 1 As shown, it includes the following steps:

[0049] Step S1: Create an environment map and initialize parameters.

[0050] This process includes the following two steps:

[0051] Step S11: Obtain the environmental map information of the robot and set the start and end points of the path planning task.

[0052] Step S12: Model the input environment. First, construct a binary map to represent obstacles and passable areas. On this basis, construct a three-layer passability semantic mesh map so that the environment not only has geometric occupancy information, but also can express passable areas in tracked mode, passable areas in hexapod mode, and areas that are impassable in both modes, thus providing a unified environmental expression for subsequent path semantic processing.

[0053] In actual navigation, hexa-legged tracked hybrid robots face not only the simple problem of determining spatial accessibility, but also the need to further differentiate the adaptation relationships between tracked and hexa-legged modes in different areas. This differentiation, in turn, supports subsequent decision-making processes such as path segmentation, mode switching, and local corrections. Therefore, if a binary occupancy grid is still used as the sole environmental description method (e.g., ...), ... Figure 2As shown in the figure, although it can reflect the spatial distribution of obstacles, it is difficult to directly express the differences in the mobility of composite robots in different terrain areas, and it is also difficult to provide a unified basis for subsequent path semantic annotation and pattern decision-making. Based on this, the present invention constructs a three-layer mobility semantic grid on the basis of a binary occupancy grid, and performs semantic annotation on the global path, so that the path includes mobility attribute information in addition to geometric information.

[0054] Specifically, based on the gridded representation of the environment, this invention divides spatial regions into three categories according to the accessibility of the robot in different motion modes: First, tracked areas are accessible, meaning areas the robot can directly and continuously traverse in tracked mode. Second, areas are inaccessible to tracked robots but accessible in hexapod mode; these areas typically correspond to low obstacles, local ditches, or complex terrain requiring elevation for crossing. Third, areas are inaccessible in both modes, representing obstacles or hazardous areas that the robot must strictly avoid. Through this division, the environmental representation is expanded from a binary occupancy relationship to a semantic grid that includes mode accessibility, such as... Figure 3 As shown.

[0055] Step S2: Generate a global path based on boot enhancement ( ).

[0056] After completing the environmental modeling, an improved approach was adopted. The algorithm performs a global path search to generate an initial global path that satisfies the collision-free constraint. The attraction and repulsion terms during the expansion process can be obtained from equations (4) and (5), respectively. Equations (6) to (8) are then combined to improve the artificial potential field, enhancing the guidance capability of the random tree towards the target direction and mitigating the impact of local minima and target inaccessibility. Simultaneously, equation (9) allows for adaptive adjustment of the expansion step size based on the distance between the current node and the nearest obstacle, thereby generating the initial global path.

[0057] for The core idea of ​​the algorithm is to gradually expand the search tree in the state space through random sampling and collision detection, without having to fully discretize the feasible space or construct a global graph structure in advance. The algorithm has the ability to handle obstacle avoidance and differential constraints, and therefore can be regarded as an open-loop trajectory generation method applicable to nonlinear systems with state constraints.

[0058] like Figure 4 As shown, The algorithm's implementation process is mainly based on the search space. Gradually expand to build a search tree until a path is formed from the starting point To the target point The feasible path. In each iteration, the algorithm... A state point is generated by random sampling, which is the random sampling point. and in the current search tree Find the node that is closest to it. Subsequently, from Along the direction The direction, with a fixed step size Generate the original new node .like and If no collision occurs between the paths, then... Add to search tree If the path is not feasible, discard the point and resample, repeating the above process until the search tree is reached. Successfully connected the starting point and the target point. The flowchart of the algorithm is as follows Figure 5 As shown.

[0059] This invention will improve The algorithm is named algorithm.

[0060] Step S2 specifically includes the following two steps:

[0061] Step S21: Introduce an improved artificial potential field guidance strategy.

[0062] Regarding the determination of the expansion direction, Relying primarily on random sampling results, node growth is highly unpredictable and prone to generating numerous invalid expansions. To address this issue, this invention introduces an improved artificial potential field model during tree growth. This model incorporates the attraction of the target point, the repulsion of nearby obstacles, and the guiding effect of random sampling points into the calculation of the expansion direction, using the direction of the combined force of these three factors as the basis for generating new nodes. This approach enhances the tendency to grow towards the target region while retaining the ability for random exploration, helping to reduce the proportion of invalid expansions and improve search efficiency. Considering the problems of local minima and target unreachability inherent in the artificial potential field method, this invention further improves the repulsion term to enhance the numerical stability and algorithm robustness during the search process.

[0063] In the original Based on this, by creating new nodes An improved artificial potential field (APF) method is added to guide the search speed of the random-growing tree by adjusting the generation direction. The artificial potential field method includes a gravity function. and repulsive function Assuming These are the robot's sampling points. The mathematical expression of the artificial potential field method is as follows:

[0064] (1)

[0065] (2)

[0066] (3)

[0067] in, and These are the gravitational field gain constant and the repulsive field gain constant, respectively. yes The maximum safe distance from the nearest obstacle area. , and Representing the nearest node Target Node and nearest obstacles The location. yes and The Euclidean distance between them yes and The distance between them. For gravity Repulsive force The combined force.

[0068] This invention is in The algorithm incorporates an improved artificial potential field method, where the target point... and random sampling points For the nearest node Applying gravity, obstacles For the nearest node When a repulsive force is applied, the direction of the resultant force of the three forces is the new node. The generation direction is determined by the algorithm growing along the resultant force direction using an adaptively adjusted step size. It should be noted that, It is the original algorithm Along random sampling points The new node obtained by directional expansion, and It is a new node obtained by expanding along the direction of the resultant force. Nearest node. The force diagram is as follows Figure 6 As shown. and These are random sampling points and target point For the nearest node Gravity, It is an obstacle For the nearest node The repulsive force. The new node is obtained according to the parallelogram law. The direction of expansion, i.e., the resultant force The direction. With most The method of this invention differs from the method of equal step size expansion. From random components (referring to random sampling points) For the nearest node The gravitational force and potential field components (referring to the target point) For the nearest node Gravity and obstacles For the nearest node The direction of growth is determined by the repulsive force (both the net force and the distance) and the magnitude of the net force, while the distance is determined by the adaptive step size adjustment function. When a node is close to an obstacle, the direction of the potential field component is mainly determined by the repulsive force, resulting in a smaller offset towards the obstacle region, which is beneficial for the random tree to move away from the obstacle. Conversely, when a node is far from an obstacle... At this time, the direction of the potential field components is mainly determined by gravity, resulting in a large offset towards the target region, which is beneficial for quickly approaching the target.

[0069] However, when the robot is about to approach the target, if there are obstacles near the target point and the repulsive force on the robot is greater than the attractive force, it will be difficult for the robot to approach the target, resulting in the target being unreachable. Figure 7 It can be seen that, due to the significant influence of the repulsive force from the obstacles surrounding the target point, the newly generated... Instead, they are somewhat distant. In fact, because Size with and The distance gradually becomes very small, if The location is not ideal. It will grow in a direction that is roughly repulsive.

[0070] To address the above problems, this invention proposes an improved APF function, the details of which are as follows:

[0071] (4)

[0072] (5)

[0073] When the robot does not have a goal unreachability problem, the gravitational field function is shown in equation (4), and the repulsive field function is shown in equation (5) to reduce the probability of local minima. When the robot has a goal unreachability problem, the new repulsive function expression is as follows:

[0074] (6)

[0075] (7)

[0076] (8)

[0077] and These are two unit vectors, representing the closest points. Towards the target point and obstacles The direction vector. The improved repulsive field function, namely formulas (5) and (6), incorporates the distance factor between the robot and the target point. This reduces the attraction and repulsion forces experienced by the robot to a certain extent during the process of moving towards the target. Only when the robot reaches the target point will the attraction and repulsion forces be reduced to zero, that is, the target point becomes the point with the minimum potential energy, thereby solving the problem of the target being unreachable. This represents the repulsive force component generated by the obstacle distance term, and its direction of action is the direction in which the robot moves away from the obstacle; This represents the guiding component generated by the target distance correction term, and its direction of action is the direction in which the robot is moving towards the target point. By introducing... The improved repulsive field gradually reduces the repulsive force as the robot approaches the target point, thereby reducing the probability of the target being unreachable.

[0078] Step S22: Perform adaptive step size adjustment.

[0079] Algorithms typically use a fixed step size for node expansion. While this approach is simple to implement, it lacks sufficient flexibility when dealing with different spatial structures, easily leading to low search efficiency in open areas and failure to expand in narrow areas, such as... Figure 8 As shown in the figure, the red step length in (a) represents expansion failure. To address this, the present invention constructs an adaptive step length adjustment based on the relative distance between the current node and surrounding obstacles, enabling the node expansion step length to adapt to local spatial constraints, thereby improving search efficiency and mobility in complex and restricted areas. The adaptive step length adjustment method uses an improved Sigmoid function as the step length adjustment model, and dynamically adjusts the expansion step length in real time based on the distance between the current expansion node and the nearest obstacle, as shown in equation (9).

[0080] (9)

[0081] in, This represents the distance between the node and the nearest obstacle. Indicates the expansion step size. When the input variable... When approaching 0, expand the step size. Approaching 0.5, when When it approaches infinity, Approaching 2. Based on this characteristic, when the extended node is in a complex and narrow space... When the step size is small, the expansion step size is automatically reduced, down to a minimum of 0.5 times the fixed step size, thus enabling finer-grained path search and improving passability in narrow spaces. Conversely, when the expansion node is located in an open area, the expansion step size is increased, up to a maximum of twice the fixed step size, improving overall search efficiency. Figure 8 As shown in Figure (b).

[0082] Step S3: Perform semantic annotation and segmentation on the path.

[0083] After step S2, an initial global path satisfying collision-free constraints is obtained. To adapt to the mobility characteristics of the hexa-legged tracked hybrid robot, the initial global path obtained in step S2 is mapped onto a three-layer mobility semantic mesh. Each path point on the path is assigned a corresponding mobility semantic label, and the path is structurally segmented according to label continuity to obtain tracked segments, hexa-legged segments, and corresponding mode-switching transition intervals. Furthermore, the equivalent length of the hexa-legged segment is calculated, providing a length basis for subsequent time cost evaluation of traversal schemes. Specifically, this step includes the following steps:

[0084] Step S31: Traverse the global path points and assign them access labels. Let the global path be... For each waypoint In the three-layer accessibility semantic grid, query the raster cell where it is located and assign it a access label. :like If it lands in an area accessible by tracks, then If it falls into an area where tracks cannot pass but six legs can, then If it falls into an area where neither mode is feasible, then .

[0085] Step S32: Divide the path into continuous segments and generate mode switching intervals. After obtaining the path point label sequence, divide the global path into several continuous segments according to label continuity. These correspond to the tracked section and the six-legged section, respectively. For each six-legged section... =[ , Furthermore, a mode switching interval is generated: pre-switching is set near the entry and exit boundaries, transforming the switching trigger from a single-point event to an interval event, thereby improving the fault tolerance and stability in the implementation. Simultaneously, the equivalent length of the hexapod segment... Defined as:

[0086] (10)

[0087] in, This is used to accumulate the arc length along the path. This length will be used to calculate the time cost of the crossing option and compared with the local cost of the bypass option. Path length together constitutes the basis for strategy judgment. , These are the two ends of the six-legged segment, that is, the starting point and the ending point.

[0088] It should be noted that the invention mentioned This refers to the A-star algorithm as a global graph search algorithm, and its local... This refers to path search performed within a local scope under the same A-star algorithm framework; both are the same algorithm, differing only in their application scope (global / local). Local Path refers to using local The detour path segment generated by the algorithm, that is, within the local area of ​​the section to be corrected, is obtained by... The algorithm-planned track bypass path used to replace the original six-legged crossing section.

[0089] Based on label continuity, the global path can be divided into tracked segments and hexapod segments. While retaining the original connectivity, the segmented path adds segment accessibility information, providing a basis for subsequent mode switching and local segment identification. A schematic diagram of path mapping and segmentation is shown below. Figure 9 As shown.

[0090] Step S4: Determine if a local correction is triggered. Traverse the semantically segmented path to identify areas requiring further decision-making. If a six-legged passage appears in the path, and a tracked passage exists nearby for detours, trigger the subsequent time cost evaluation process. If the triggering condition is not met, maintain the original path.

[0091] After completing the semantic representation of the environment and the semanticization of the global path, an initial global path with accessibility attribute markers can be obtained, and the segments corresponding to the tracked and hexapod modes can be identified. However, relying solely on the offline-generated semantic global path cannot guarantee that the access strategy for certain segments will always be reasonable. This is because when low obstacles, ditches, and detours coexist, directly entering the hexapod segment along the original path may not necessarily be the optimal solution in terms of overall time cost. Therefore, after generating the initial global path and completing semantic segmentation, it is necessary to perform path-level correction and mode decision-making for local segments based on time cost. Without changing the overall direction of the global path, the crossing and detour judgments are made for local segments, and when the detour solution has a lower time cost, the original path is partially replaced by an alternative path segment.

[0092] Step S5: Evaluate the time cost of crossing and detouring. For the area after the trigger, calculate the time cost of the six-legged crossing scheme and the tracked detouring scheme respectively.

[0093] For the traversal method, the robot must first switch from tracked mode to hexapod mode, traverse low obstacles or ditches in hexapod mode, and then revert to tracked mode after passing through. The time cost of the traversal method... for:

[0094] (11)

[0095] in, This refers to the time consumed during a single mode switch. The equivalent length of the global path for areas where low obstacles or ravines can be crossed. For six-legged mode speed, ∈{1, 2} represents the number of switching operations; if the tracks are restored immediately after crossing, then... =2, otherwise 1.

[0096] The above results take into account the time consumed by mode switching, the movement time of the six-legged mode through the section, and the number of switching operations.

[0097] For the detour plan, the track mode is kept unchanged, and a planning domain is constructed near the starting position of the section to be corrected. In this embodiment, the method for constructing the planning domain is as follows: starting from the starting position of the section to be corrected... Centered on, with A circular region with radius is used as the local planning domain, where Determined adaptively based on environmental complexity , and The environmental complexity was statistically analyzed using 10m and 50m as values. Centered on The percentage of obstacle grids within a circular area with a radius of 20m To characterize, when The area was determined to be open at that time, and was taken. ,when When it is determined to be a region of medium complexity, take ,when When it is determined to be a narrow area, take Perform local operations under the constraints of the track's traversable space. The algorithm generates the starting position of the segment to be corrected. After reaching the low obstacle, the target point is... path segment The time cost of the detour option. for:

[0098] (12)

[0099] in, For local Algorithm generates detour segments Length, This refers to the track speed.

[0100] To reduce cost fluctuations caused by sampling randomness, local [measures] can be implemented. Set a maximum planning time limit, and select the shortest feasible path within the maximum time limit. If a feasible path cannot be found within the maximum time limit, the detour is deemed infeasible, and the process reverts to a crossing option. Therefore, the time cost of the detour option is determined locally. The length of the detour path and the speed of movement in track mode are jointly determined.

[0101] After obtaining the time costs of the two types of solutions, a mode decision is made based on the discrimination rule with hysteresis margin. Let... Hysteresis margin ( If ), then the following decision criteria can be adopted: If If the vehicle chooses to detour and maintain tracked mode, it will switch to six-legged traverse mode. This is used to suppress decision jitter and frequent mode switching caused by local environmental perturbations and path length estimation fluctuations in critical conditions. The purpose of this approach is to transform mode decision-making from a simple instantaneous comparison into a robust discrimination process with a certain stability margin, thereby improving the reliability of local correction triggering. In this case, we can directly compare the results of equation (11) and equation (12) to determine whether it is more reasonable to cross over or detour in the current area.

[0102] If the detour is more cost-effective, proceed to step S6; if the crossing is more cost-effective, retain the original path.

[0103] Step S6: Generate local Detour path segment. When the time cost assessment results indicate that tracked detour is superior to hexapod crossing, a local detour is invoked within the critical region neighborhood. The algorithm generates detour path segments. Simultaneously, candidate sub-target points are determined by advancing along the original global path using equation (13), and the final reconnection location is determined after combining semantic attributes, collision states, and reachability filtering, enabling the local detour segment to avoid the original local segment and rejoin the original global path. In other words, to ensure a smooth connection between the local detour segment and the global path, and to prevent sub-target points from falling into low obstacles, ditches, or other complex areas that could lead to local planning failure, this invention adopts a sub-target point selection strategy based on forward advancement of the global path.

[0104] Let the global path be The starting position of the section to be corrected is First of all, Search and Recent path index and advance the distance along the path. Candidate points were obtained :

[0105] (13)

[0106] in, , The value can be adjusted according to the degree of openness of the environment: larger values ​​are used in open areas to reduce the frequency of local replanning, and smaller values ​​are used in narrow areas to improve feasibility. A set of pathpoints representing a global path.

[0107] After obtaining candidate sub-target points, their semantic attributes, collision states, and reachability must be considered for further filtering. Only points located within the track-accessible area and locally... A point is only considered a final sub-target point when a feasible path to it can be generated within the search area of ​​the current local segment. This is to ensure that the detour path can both cross the original local segment and naturally rejoin the global path in space, avoiding local planning failures or abrupt path splicing due to improper target selection.

[0108] Step S7: Global path concatenation. This involves concatenating the local paths generated in step S6. The detour path segment is concatenated with the remaining part of the original global path after the re-entry point. In other words, when the determination result is detour, a local correction method is used to concatenate the alternative path segment with the remaining part of the original global path after the sub-target point to form a corrected global path.

[0109] This operation only replaces a local segment, without changing the overall direction and connectivity of the original global path. Compared to replanning the entire path, this method only processes data within a local area. To further ensure the robustness of the local correction process, a corresponding rollback mechanism is also implemented: if the local... If a feasible path to the sub-target point cannot be generated within the time limit, the detour is deemed infeasible, and the crossing path corresponding to the original local segment is retained.

[0110] Step S8: Replace the six-legged crossing segment with the tracked bypass segment. When a local bypass solution is better, replace the corresponding six-legged crossing segment in the original path with the corrected path obtained in Step S7, and output the final planning result. If the local... If a feasible detour path cannot be generated under the given conditions, or if the comparison results show that the crossing scheme is better, then the original path is retained unchanged.

[0111] To verify the effectiveness of this invention, simulation experiments were conducted under unified hardware and software conditions. The simulation program was written in Python and ran in the PyCharm integrated development environment. The computing platform was a computer equipped with an Intel Core i7-13700HX CPU and an NVIDIA RTX 4050 GPU.

[0112] Hexa-legged tracked composite robot entity as Figure 10 As shown, the relevant simulation parameters were set based on the actual parameters of the hexa-legged tracked composite firefighting robot. The robot's dimensions were 532 cm × 226 cm in tracked mode and 532 cm × 553 cm in hexa-legged mode. The baseline speed was set to 0.8 m / s in tracked mode and 0.4 m / s in hexa-legged mode. The threshold for crossing low obstacles was set to 0.50 m; obstacles higher than 0.50 m were considered high obstacles and could not be crossed in either tracked or hexa-legged mode. In other words, obstacles between 0 and 0.5 m in height could be crossed by the hexa-legged robot. The ditch terrain was set to a depth of 2 m and a width of 2.4 m. The mode switching waiting time was uniformly set to 10 s. The robot was required to complete the mode switch 6 m in front of the terrain to be switched and switch again 6 m behind the obstacle after crossing it, allowing for a unified comparison of the time costs of crossing and detouring strategies. Experimental scenarios included random environments and simulated factory environments, used to examine the adaptability of the proposed method under discrete obstacle distribution conditions and enhanced structural constraints, respectively. The random environment map is 300 m × 200 m in size, such as... Figure 11 As shown, the simulated factory environment map is 400 m × 200 m in size. Figure 12 As shown.

[0113] To comprehensively evaluate the performance of the proposed method, this invention sets up two sets of control experiments. The first set, under the same compounding mode configuration, compares... Algorithms and The first set of algorithms was used to compare the differences between the two methods in terms of path length, planning time, number of samples, and number of turning points. The first set of experiments was conducted independently in both random and simulated factory environments to compare performance differences under different scenario conditions. The second set of experiments, using the same environment map and the same global path planning algorithm, compared the pure tracked mode and the hexa-tracked hybrid mode to analyze the impact of the hybrid motion mode on the global path search space and path results. Except for the passability boundary, all other experimental conditions remained consistent. The pure tracked mode could not pass through the yellow low obstacles and blue ditch areas, while the hybrid mode could switch to hexa-track mode to cross them; therefore, the passable space ranges for the two modes differed.

[0114] To ensure the comparability of the results, all experiments employed consistent semantic map construction rules, motion parameters, and mode switching strategies. Evaluation metrics included path length, planning time, number of samples, and number of turning points, comprehensively reflecting path geometric quality, actual traffic efficiency, and algorithm computational burden. Subsequent analyses will combine simulation trajectory graphs and statistical results to further examine the performance differences between different methods and modes.

[0115] To verify the role of time cost discrimination and local correction mechanisms in local segment decision-making, this invention conducted simulation analysis on a 300m×200m random environment map. The results are as follows: Figure 13 , 14 As shown in Table 1, the environment is represented by a three-layer drivability semantic map. Black areas represent obstacles impassable in both modes; yellow areas represent low obstacles that cannot be crossed in tracked mode but can be crossed in hexapod mode; and blue areas represent ravine terrain impassable in tracked mode but passable in hexapod mode. Along... The generated initial global path identified 5 local segments requiring time cost assessment, i.e., 5 decision points, namely: Figure 13 The five points a, b, c, d, and e are defined. For each section, the estimated time for the six-legged crossing and the tracked detour are calculated, and based on this, it is determined whether the original path should be partially replaced.

[0116] Figure 13 , 14 show, The generated initial global path intersects with low obstacles or ravines in several local sections, indicating that the initial path obtained solely based on geometric reachability does not necessarily correspond to a more efficient solution in terms of actual travel time. After local correction, the path adjustment is mainly concentrated near local sections, and the corrected path still maintains the overall orientation of the original global path, demonstrating that this mechanism performs targeted optimization of local sections while maintaining the basic stability of the global structure. Figure 13 , 14 In the table, the modified candidate segment represents the candidate path for detour when facing key points. Table 1 further provides the crossing time, detour time, crossing distance, and detour distance for the five decision points. The data in the table shows that the crossing distance for all five decision points is less than the detour distance, but the crossing time is not always shorter. This indicates that for hexapod tracked hybrid robots, the strategy selection for local segments cannot be based solely on the local path length; it is also necessary to comprehensively consider the lower speed of the hexapod mode and the additional time consumption caused by mode switching.

[0117] Table 1 Simulation data of decision points for the composite robot

[0118]

[0119] Specifically, the detour distances at decision points a, b, and c are 34.75 m, 36.98 m, and 42.60 m, respectively, all slightly longer than the corresponding crossing distances of 31.66 m, 33.43 m, and 36.33 m. However, their detour times are 43.44 s, 46.22 s, and 53.25 s, respectively, significantly shorter than the corresponding crossing times of 99.14 s, 93.57 s, and 110.82 s. Therefore, these three local sections utilize tracked vehicle detours, and the original paths are partially replaced. The detour distances at decision points d and e are 180.73 m and 159.96 m respectively, which are much larger than the crossing distances of 25.41 m and 22.26 m. The corresponding detour times also increase to 225.91 s and 199.95 s respectively, which are significantly higher than the crossing times of 73.51 s and 75.65 s. Therefore, the original crossing paths are retained for these two sections.

[0120] comprehensive Figure 13 , 14 As shown in Table 1, the local correction mechanism can differentiate time costs for different local segments, making local corrections to the original path when the estimated detour time is lower, and maintaining the original path when a more suitable crossing is chosen. This result indicates that in path planning for hexa-legged tracked hybrid robots, relying solely on the initial global path is insufficient to guarantee the rationality of local segment decisions; therefore, introducing a local correction mechanism is necessary. This mechanism can improve the consistency between local segment strategy selection and actual traffic efficiency without altering the overall global path, thus making the planning results more consistent with the traffic characteristics of hexa-legged tracked hybrid robots.

[0121] To analyze the performance of different path planning algorithms on the same hexa-track composite mode, this invention, under the condition of consistent robot motion mode, motion parameters, and experimental environment, [conducts / tests]. Two existing improved method( and )and Comparative experiments were conducted, and the algorithm's performance was analyzed by combining paths and statistical results from two types of maps. The algorithm's performance was evaluated using four metrics: path length, planning time, number of samples, and number of turning points on the path. The number of samples was defined as the total number of all sampling attempts, including samples landing on obstacles and samples deemed infeasible during collision detection; therefore, it is normal for the number of samples to exceed the actual number of nodes in the random tree. The definition of a turning point is as follows: This invention uses the heading change angle between adjacent path segments as a discrete turning metric. When this turning angle exceeds a threshold... When this happens, the corresponding path point is determined to be a turning point. The threshold can effectively filter out small disturbances caused by sampling randomness, and also capture those broken-line turns that affect the robot's motion comfort, thus more reasonably reflecting the overall smoothness of the path.

[0122] Figure 15 As shown in Table 2, under random environment map conditions, The algorithm outperforms other algorithms in four metrics: global path length, planning time, number of samples, and number of inflection points. algorithm. The generated global planned path length is 306.28 m, the planning time is 511.9 ms, the number of samplings is 341, and the number of turning points is 56. The generated global planned path length is 262.86 m, the planning time is 184.5 ms, the number of samplings is 231, and the number of turning points is 4. The generated global planned path length is 282.73 m, the planning time is 958.6 ms, the number of samplings is 526, and the number of turning points is 3. The generated global planned path length is 276.65m, the planning time is 778.7 ms, the number of samples is 376, and the number of turning points is 6. Combined with... Figure 15 As can be seen, the improved algorithm generates a more direct path with fewer polygonal turns. For a hexa-tracked hybrid robot, this means that in long-distance travel sections dominated by tracked mode, the geometric redundancy caused by frequent turns can be reduced. When approaching key areas such as low obstacles and ditches, the clearer path direction also facilitates subsequent path semantic segmentation and mode switching location determination. Therefore, the results in the random environment map demonstrate that this improved method not only improves the geometric quality of the initial global path but also makes the resulting path more suitable for the travel requirements of the hexa-tracked hybrid robot.

[0123] Table 2 Simulation data of different algorithms for the same composite mode in random environment maps

[0124]

[0125] Figure 16 As shown in Table 3, under the condition of the factory area environment map, the gap between the two algorithms on four indicators further widened. The generated global path length is 452.12 m, the planning time is 2826.6 ms, the number of samplings is 2420, and the number of turning points is 81. The generated global path length is 383.51 m, the planning time is 1234.4 ms, the number of samplings is 542, and the number of turning points is 3. The generated global path length is 368.79 m, the planning time is 33450.4 ms, the number of samplings is 2864, and the number of turning points is 1. The generated global path length is 358.90 m, the planning time is 4594.8 ms, the number of samples is 675, and the number of turning points is 4. Combined with... Figure 16 It is evident that in factory settings with larger environmental scales and more continuous constraints, It still exhibits strong random expansion characteristics, resulting in a large number of invalid samples and obvious path detours. and Although the path length is shorter, the planning time and number of samples are higher. In contrast, the improved algorithm, through adaptive step size and improved artificial potential field guidance, reduces blind expansion and makes the overall path direction closer to the target direction. This shows that under the same composite mode conditions, the algorithm improvement not only improves search efficiency but also enhances the compactness and smoothness of the initial global path.

[0126] Table 3. Simulation data of different algorithms for the same composite mode of the factory environment map

[0127]

[0128] The experimental results from both the random environment map and the factory environment map show that, under the same hexapod tracked composite robot mode conditions, The algorithm outperforms the basic algorithm in four metrics: global planning path length, planning time, number of samplings, and number of turning points. More importantly, this advantage aligns well with the platform characteristics of hexa-tracked hybrid robots. Hexa-tracked hybrid robots are not simply mobile platforms pursuing the geometric shortest path; they need to simultaneously balance efficient continuous propulsion in tracked mode and the ability to traverse complex terrain in hexa-legged mode. Therefore, their global paths must not only be reachable but also as smooth, compact, and clearly structured as possible. The improved algorithm proposed in this invention demonstrates better adaptability in these aspects, generating shorter, smoother paths with fewer turns, reducing geometric redundancy and turning requirements in tracked-dominant sections. Furthermore, its lower sampling cost and faster planning process are more conducive to providing stable and reliable global paths for hybrid robots in complex environments.

[0129] To analyze the impact of different motion modes on path planning results and actual mobility, this invention conducted comparative experiments on pure tracked mode, pure hexapod mode, and hexapod-tracked hybrid mode while maintaining consistent environmental maps, start-end locations, and path planning methods. The applicability of the two motion modes was analyzed by combining path comparison results and statistical indicators from the two types of maps. Evaluation indicators still include global planning path length, planning time, number of samplings, and number of turning points, reflecting path compactness, planning efficiency, search cost, and path smoothness, respectively. For hexapod-tracked hybrid robots, these indicators not only reflect the performance of the planning algorithm itself but also the degree of matching between the path results and the robot's hybrid mobility.

[0130] Depend on Figure 17 As shown in Table 4, in a randomized map environment, the hexa-tracked composite mode performs worse than both the pure tracked and pure hexa-tracked modes in four key metrics: global path length, planning time, number of samplings, and number of turning points. In the pure tracked mode, these metrics are 357.74 m, 2486.3 ms, 6966 samplings, and 7 turning points, respectively. In the pure hexa-tracked mode, these metrics are 293.83 m, 442.7 ms, 582 samplings, and 5 turning points, respectively. In the hexa-tracked composite mode, these metrics are 262.86 m, 184.5 ms, 231 samplings, and 4 turning points, respectively. Combined with the path map, it can be seen that the pure tracked mode can only search for paths within the traversable area of ​​the track. When encountering low obstacles and ditches, it needs to avoid them in advance, resulting in a narrower range of selectable paths and more obvious path detours. In contrast, the composite mode allows the robot to use the hexa-tracked mode to traverse local sections, so that low obstacles and ditches are no longer considered as obstacles that must be avoided, thus expanding the effective traversable space available for planning. Therefore, the difference between the two modes in a random environment map is essentially a change in the search space and the resulting difference in path results.

[0131] Table 4. Simulation data of the same algorithm under different motion modes in random environment maps.

[0132]

[0133] Depend on Figure 18As shown in Table 5, under the factory environment map, the gap between the two movement modes in the four indicators further widens. In the pure tracked mode, the global path length is 495.07m, the planning time is 4379.7ms, the number of samples is 14127, and the number of turning points is 7. In the pure hexapod mode, the above indicators are 372.06m, 2896.5ms, 220 samples, and 6 turning points, respectively. In the hexapod-tracked hybrid mode, these indicators are 383.51m, 1234.4ms, 542 samples, and 3 turning points, respectively. Compared with the random environment map, the factory environment map has a larger environmental scale and more continuous constraints. The pure tracked mode is more easily limited by obstacle distribution and traversable bandwidth, and can only make long-distance detours along the periphery or narrow passages, resulting in a decrease in search space utilization and an increase in the number of samples. The hybrid mode, on the other hand, allows the robot to cross low obstacles and ditch areas, expanding the traversable area in the planning stage, and thus can search for paths to the target in a larger space. This indicates that, under the same algorithm conditions, differences in motion patterns not only affect the terrain boundaries that the robot can pass through, but also change the way the search space is expressed in the global path planning stage, and further affect the path generation results.

[0134] Table 5. Simulation data of the same algorithm for different motion modes in the factory environment map.

[0135]

[0136] Finally, it is necessary to state that the above embodiments are only used to further illustrate the technical solution of the present invention in detail, and should not be construed as limiting the scope of protection of the present invention. Any non-essential improvements and adjustments made by those skilled in the art based on the above content of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A path planning method for a hexa-legged tracked composite robot, characterized in that, Includes the following steps: Step S1: Create an environment map, construct a binary map and a three-layer navigability semantic grid; Step S2: Adopt the improved The algorithm generates an initial global path, and the improvements include the introduction of an improved artificial potential field guidance and adaptive step size adjustment; The improved artificial potential field guidance involves incorporating the target point's gravity, the random sampling point's gravity, and the obstacle's repulsive force into the expansion direction calculation, using the resultant force direction as the new node generation direction; when the target is unreachable, the repulsive force function is decomposed into... and Two components, This represents the repulsive force component generated by the obstacle distance term, and its direction of action is the direction in which the robot moves away from the obstacle. The guiding component generated by the target distance correction term acts in the direction the robot is moving towards the target point. The adaptive step size adjustment involves dynamically adjusting the step size based on the distance between the current node and the nearest obstacle, reducing the step size in narrow areas and increasing it in open areas. This dynamic adjustment uses a Sigmoid function-based adjustment model. The step size range includes 0.5 times to 2 times the fixed step size; Step S3: Map the initial global path to the three-layer drivability semantic mesh, perform semantic annotation and segmentation on the path to obtain tracked segments and hexapod segments; the semantic annotation and segmentation includes: assigning track, leg, or blocked labels to path points, dividing the global path into several continuous segments according to the continuity of the labels, the segment where the track label is located is the tracked drivable area, the segment where the leg label is located is the hexapod drivable area, and the segment where the blocked label is located is the area that is impassable in both modes, and for each hexapod segment, set a pre-switching near the entry and exit boundaries to change the switching trigger from a single-point event to an interval event; the equivalent length of the hexapod segment is the difference between the cumulative arc lengths of the two ends of the hexapod segment along the path; Step S4: For the six-legged segment, determine whether there is a tracked detour path. If so, proceed to time cost evaluation; otherwise, maintain the original path. Step S5: Calculate the time cost of six-legged crossing and tracked detour respectively, and decide whether to cross or detour based on the hysteresis margin discrimination rule; the time cost of the six-legged crossing scheme. Time cost of track bypass scheme for: ； ； in, This refers to the time consumed during a single mode switch. The equivalent length of the global path for areas where low obstacles or ravines can be crossed. For six-legged mode speed, ∈{1, 2} represents the number of switching operations; if the tracks are restored immediately after crossing, then... =2, otherwise 1. For local The algorithm generates the length of the detour segment. Let the track speed be denoted by the rule for determining the track hysteresis margin. As a hysteresis margin, ,like If the vehicle chooses to detour and maintain the tracked mode, it will choose to cross on six legs. In the detour scheme, the radius of the planning domain is adaptively determined based on the environmental complexity, with the starting position of the section to be corrected as the center. The environmental complexity is characterized by the proportion of obstacle grids within the preset radius centered on the starting position. Step S6: If the decision is to detour, generate a detour path segment and concatenate it with the global path, replacing the original hexapod segment; the generation of the detour path segment adopts a local approach. The algorithm plans within the passable space of the track. The splicing includes advancing along the original global path to determine candidate sub-target points, and determining the final reconnection position after combining semantic attributes, collision status, and reachability filtering, so that the local detour path segment can avoid the original local segment and rejoin the original global path. The detour path segment is spliced ​​with the remaining path of the original global path after the reconnection point to form a modified global path. The modified global path only replaces the local segment and does not change the overall direction and start-end connectivity of the original global path. If a feasible path to the sub-target point cannot be generated when generating the detour path segment, the detour is determined to be infeasible, and the crossing path corresponding to the original local segment is retained. Step S7: Output the final planned path.

Images